feat(cli): set AI_AGENT for subprocesses

[complynx]

Summary

• expose AI_AGENT=kimi to subprocesses launched from both the pip/uv and standalone binary entrypoints
• preserve an explicit non-blank value supplied by a wrapper or orchestrator
• cover missing, blank, and explicit marker behavior across both entrypoints

Why a universal marker

Tools launched by coding agents often need to adapt telemetry, UX, logging, or safety behavior based on whether they are running under an agent. Today they generally have to maintain agent-specific process-tree and environment heuristics. A shared AI_AGENT=<slug> convention lets the producer that already knows its identity declare it once, avoiding an N-by-M matrix of consumer heuristics as agents and tools multiply.

This is deliberately an advisory marker, not a security boundary. Kimi only supplies its own slug when the variable is missing or blank, so wrappers and higher-level orchestrators retain control of any explicit non-blank value.

Validation

• uv run ruff format --check src/kimi_cli/__main__.py src/kimi_cli/cli/__main__.py src/kimi_cli/utils/subprocess_env.py tests/core/test_startup_imports.py
• uv run ruff check src/kimi_cli/__main__.py src/kimi_cli/cli/__main__.py src/kimi_cli/utils/subprocess_env.py tests/core/test_startup_imports.py
• uv run pytest tests/core/test_startup_imports.py tests/utils/test_subprocess_env.py -q (14 passed)

[devin-ai-integration]

Devin Review found 1 potential issue.